Here's a completely wild hypothesis:
Some previous (really broken) system working on this data tried to write each character as UTF-8, but actually wrote only the last byte of each sequence (maybe there was a strange one-byte buffer somewhere). Alternatively, this was in UTF-8 in the past, but someone, looking at it in a different encoding, performed a search and replace to remove the 0xE2 0x80 bytes, because they clearly did not belong and did not understand that the remaining ones were special the symbol "was not what they wanted.
ASCII, , , UTF-8 .
" " (U + 2019) ’ UTF-8 0xE2 0x80 0x99. , \x99s, , , s . , 0x99.
" " (U + 201D) " UTF-8 0xE2 0x80 0x9D. 0x9D, , . ". , - - , "" 0x9D.
, , , , . UTF-8 "" , , - , .